build: delete checks for C89 standard headers

[vszakats]

Delete checks and guards for standard C89 headers and assume these are
available: stdio.h, string.h, time.h, setjmp.h, stdlib.h,
stddef.h, signal.h

Some of these we already used unconditionally, some others we only used
for feature checks.

Follow-up to 9c7165e #11918 (for stdio.h in CMake)

Closes #11940

[ncopa]

stdio.h is needed to detect fseeko, both on MacOS and Linux.

-- Looking for fseeko
-- Looking for fseeko - not found

[vszakats]

@ncop: Do you think this solution would also work?:

check_symbol_exists(fseeko         "${CURL_INCLUDES};stdio.h" HAVE_FSEEKO)
check_symbol_exists(_fseeki64      "${CURL_INCLUDES};stdio.h" HAVE__FSEEKI64)

This also self-documents that stdio.h is necessary for detection, without adding it for every other check by default.

[vszakats]

@bagder: I extended this PR with other C89 headers that we're checking for, most likely unnecessarily. I'm looking into signal.h and plan to add it shorly.

[vszakats]

ea8fbb5 (from 2008) suggests that a system called "Cell" might not actually come with a C89 signal.h. According to the original report this is CellOS for PlayStation 3. Could this be still relevant (or on other systems) to keep checking for signal.h?

[jay]

c89 4.1.2 standard headers:

         <assert.h>                 <locale.h>                 <stddef.h>
         <ctype.h>                  <math.h>                   <stdio.h>
         <errno.h>                  <setjmp.h>                 <stdlib.h>
         <float.h>                  <signal.h>                 <string.h>
         <limits.h>                 <stdarg.h>                 <time.h>

regarding HAVE__FSEEKI64 see #11944

[vszakats]

I plan to follow this up with a patch to move CMake global standard headers (string.h, stdlib.h, stddef.h) to the specific detections where they are actually needed (if they are) based on autotools logic. Also found an incomplete/broken detection in CMake and one missing compared to autotools.

[ncopa]

@ncop: Do you think this solution would also work?:

check_symbol_exists(fseeko         "${CURL_INCLUDES};stdio.h" HAVE_FSEEKO)
check_symbol_exists(_fseeki64      "${CURL_INCLUDES};stdio.h" HAVE__FSEEKI64)

This also self-documents that stdio.h is necessary for detection, without adding it for every other check by default.

Yes, I would expect that work work, and according the logs it does:

-- Looking for fseeko
-- Looking for fseeko - found

I agree that it is better this way.

[ncopa]

Also found an incomplete/broken detection in CMake and one missing compared to autotools.

I only discovered the broken test of fseeko, because I was looking at the logs. It is fragile because the fallback code also works, so a broken test can easily go undetected.

Maybe it would be an idea to add a test to CI to run both configure and cmake and compare the generated curl_config.h?

[bagder]

Maybe it would be an idea to add a test to CI to run both configure and cmake and compare the generated curl_config.h?

That's a good idea. It will of course only test the exact options used in the test and we can run a few different ones, but it should still help us.

[vszakats]

Such detection comparison test would be very useful indeed.

[MarcelRaad]

This seems to break strerror_r detection on Solaris, visible as warnings in the autobuilds since yesterday:
https://curl.se/dev/log.cgi?id=20230927074238-1773871#prob1

configure: WARNING: cannot determine strerror_r() style: edit lib/curl_config.h manually.

[vszakats]

Is it possible Solaris doesn't have a string.h? Or it something else? Is there a way to pull or view config.log of a failed build?

[vszakats]

According to this, strerror_r is declared in string.h: https://docs.oracle.com/cd/E86824_01/html/E54766/strerror-r-3c.html

[vszakats]

Possibly related (caused by my bad edit): d14089d